Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

量化后变得更慢了? #34

Closed
sonixixi opened this issue Feb 21, 2019 · 4 comments
Closed

量化后变得更慢了? #34

sonixixi opened this issue Feb 21, 2019 · 4 comments

Comments

@sonixixi
Copy link

No description provided.

@sonixixi
Copy link
Author

在 armv7 上 速度变慢了两倍

@BUG1989
Copy link
Owner

BUG1989 commented Feb 21, 2019

@stupidchenchen 用于部署的框架正在努力重构中,armv7a框架已经支持较好了,新增以下特性:

  1. 采用了权重分通道量化,精度更高;
  2. conv3x3s1采用了int8 winograd F(2,3),虽然部分尺寸还是没有float32 winograd F(2,3)快(我会竭尽全力超过它,但是都是业余时间开发,摊手.jpg),但是比目前master分支中的还是快一些;
  3. 其他尺寸卷积的int8使用im2col+sgemm,目前master中是naive c实现的;
  4. 本次重构的初衷是实现int8 without overflow,即用精度与x86 simulator完全一致,该功能已实现,精度可放心使用;
  5. arm64-v8a正在开发中……
  6. 想不起来了……

ncnn int8 new implement pr
完整的repo

@sonixixi
Copy link
Author

@BUG1989 抱紧大佬大腿

@BUG1989 BUG1989 pinned this issue Feb 25, 2019
@BUG1989 BUG1989 closed this as completed Mar 7, 2019
@BUG1989
Copy link
Owner

BUG1989 commented Apr 5, 2019

The latest pr has been done.Add the int8 winograd F(4,3) implement.
SpeedUp.....www
add int8 winograd F(4,3) with neon assembly optimization

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants